Diameter and height modeling for accurate prediction of tree size in a Douglas-fir rainforest

Abstract Modern forest inventories increasingly integrate ground and aerial datasets. Often, prediction of both tree heights from ground measurements and tree diameters from aerial point clouds is thus required. This study jointly evaluates 77 fixed-effect regression forms predicting either (1) total height or (2) diameter at breast height of individual trees. Three conifer, three broadleaved, and a group of less common tree species in North America’s central Pacific Temperate Rainforest are considered. Prediction accuracy was dominated by selection of base model form and differences between naturally regenerated and plantation stands, resulting in model efficiencies near 90% for height and 85% for diameter. Inclusion of generalizing stand structure and physiographic variables increased height model efficiency by 0.0%–1.2%, comparable to the 0.0%–0.9% increase from generalizing diameter models. This broad evaluation and selection process enables increased forest inventory accuracy and improved tree growth prediction by evaluating new and existing allometric model forms, creating or substantially revising allometric models for study area species, and establishing a basis for further model development in any forest. Generalized additive models, in particular, were preferred to nonlinear or linear regressions in 65% of species and response variable combinations, indicating opportunity to revise nonlinear regressions to better utilize their greater interpretability, quicker fitting, and rapid evaluation times.